Title

Interspecific interactions define Staphylococcus aureus infection risk during mechanical ventilation

Abbreviated Title

Respiratory Corynebacterium and Staph aureus

Authors

Brendan J. Kelly, MD, MS (1,2); Ebbing Lautenbach, MD, MPH, MSCE (1,2); Jason Roy, PhD (5); Ayannah S. Fitzgerald, BSN (3,4); Layla A. Khatib, BS (3,4); Ize Imai, MS (4); Leigh Cressman, MA (2); Frederic D. Bushman, PhD (3); Ronald G. Collman, MD (3,4)

Affiliations

1 - Division of Infectious Diseases, Department of Medicine;
2 - Department of Biostatistics, Epidemiology, and Informatics;
3 - Department of Microbiology;
4 - Division of Pulmonary, Allergy, and Critical Care, Department of Medicine, Perelman School of Medicine, University of Pennsylvania, Philadelphia, PA;
5 - and Department of Biostatistics and Epidemiology, Rutgers University, New Brunswick, NJ

Corresponding Author Contact

Brendan J. Kelly, MD, MS -

Authors’ Contributions

BJK - study design, data collection, data analysis, manuscript preparation;
EL - study design, manuscript revisions;
JR - statistical methods, manuscript revisions;
ASF - subject enrollment, specimen collection, data collection;
LAK - subject enrollment, specimen collection, data collection;
II - specimen processing, data collection;
LC - data collection;
FDB - study design, manuscript revisions;
RGC - study design, manuscript revisions

Disclosures

The authors report no relevant disclosures.

Data Availability

Processed clinical and microbiome data, as well as analysis scripts and model code are available at github.com/bjkellyio. Raw 16S rRNA gene sequence data have been made publicly available at the National Center for Biotechnology Information’s Sequence Read Archive (NCBI SRA) with BioProject ID: PRJNA529220.

Keywords

Staphylococcus aureus, Ventilator-Associated Pneumonia, Microbiome, Corynebacterium


Manuscript

Abstract

  • Question Addressed: Intranasal colonization with Staphylococcus aureus ( Sa ) is a marker for increased risk of invasive Sa infection. Prior studies in healthy subjects suggest other nasal bacteria may interact with Sa and impact risk of colonization, but their impact on invasive disease is unknown. We sought to define the relationship between intranasal and respiratory tract microbiota among critically ill subjects dependent on mechanical ventilation and identify features that impact risk of Sa lower respiratory tract infection (LRTI).
  • Patients and Methods: We enrolled two prospective cohorts (pilot n = 14; validation n = 76) of subjects dependent on mechanical ventilation in the medical intensive care unit of an academic medical center. We collected longitudinal anterior nares (AN) and endotracheal aspirate (ET) specimens to analyze bacterial community composition by 16S ribosomal RNA gene sequencing.
  • Results: Three intranasal Sa amplicon sequence variants (ASVs), which were identified across subjects, cohorts, and sequencing runs, were associated with increased Sa abundance in both AN and ET sites, as well as increased risk for Sa LRTI. Of 597 non- Sa intranasal ASVs, 11 were identified as protective against Sa LRTI, even accounting for the presence of Sa ASVs and prior antibiotic exposure. 8 of 11 (72.7%) protective ASVs were Corynebacterium species.
  • Conclusions: During critical illness and mechanical ventilation, intranasal microbiome features define the risk of Sa ventilator-associated LRTI. Intranasal Sa ASVs associated with high lower respiratory abundance increase risk, but interspecific interactions largely driven by Corynebacterium decrease the risk of invasive disease posed by Sa nasal colonization.

Introduction:

Staphylococcus aureus (Sa) colonizes human skin and mucosa. The anterior nares is the most common site of colonization, but Sa may also reside in the axillae, perineum, and pharynx. Approximately 30% of humans demonstrate persistent intranasal Sa colonization, and intranasal Sa colonization has long been recognized as a marker of risk for invasive infection [1]. Patients admitted to hospitals with intranasal Sa colonization have approximately three-fold increased risk of non-surgical healthcare-associated Sa infection [2–6]. Sa accounts for approximately 20% of cases of ventilator-associated lower respiratory tract infection (VA-LRTI), including pneumonia and tracheobronchitis, with an incidence of approximately 4.5 cases of Sa LRTI per 1000 mechanical-ventilation days [7–10]. What distinguishes patients with Sa colonization who do or do not develop invasive disease is a gap in current knowledge important for clinical prediction and the design of novel prevention and treatment interventions.

Among healthy adults, interactions between Sa and Corynebacterium species have been shown impact intranasal Sa colonization [11–14], and in vitro studies demonstrate Corynebacterium species are capable of attenuating Sa virulence [15]. However, the impact of interactions among members of the intranasal microbiome on respiratory Sa colonization in critically ill patients, an especially vulnerable population, and particularly on Sa ventilator-associated LRTI, remains undefined [16].

We sequentially enrolled pilot then validation cohorts of critically ill adults who were newly-intubated and dependent on mechanical ventilation. We obtained anterior nares (AN) and endotracheal aspirate (ET) specimens on enrollment and longitudinally for the duration of their critical illness, and we ascertained Sa LRTI outcomes. We first developed models for the impact of nasal Sa strains and interspecific interactions on risk of Sa ventilator-associated LRTI, then validated these models. We found that intranasal colonization with specific Sa sequence variants increases Sa LRTI risk, but that the risk of Sa LRTI associated with intranasal Sa colonization is reduced when Sa is accompanied by specific Corynebacterium sequence variants.


Materials and Methods:

  • Study Design and Setting: We performed a prospective cohort study, enrolling subjects from the medical intensive care unit of the Hospital of the University of Pennsylvania (HUP), an academic medical center in Philadelphia, PA, USA, within 24 hours of their starting mechanical ventilation. Initial AN and ET specimens were obtained at enrollment, and specimen collection was repeated at 48- to 72-hour intervals thereafter for the duration of mechanical ventilation. Informed consent was obtained from subjects or their surrogates. The protocols were reviewed and approved by the University of Pennsylvania IRB (protocols #817706 and #823392). After an initial, pilot cohort of 14 subjects demonstrated the importance of intranasal Sa ASVs to LRTI risk and suggested a possible protective role for intranasal Corynebacterium jeikeium, a validation cohort of 76 subjects was enrolled to further interrogate these findings.

  • Study Population: Subjects were eligible for inclusion with (1) age >= 18 years, (2) new onset of dependence on intubation and mechanical ventilation, and (3) an anterior nares swab that could be collected within 24 hours of starting mechanical ventilation. Potentially eligible subjects were excluded if they had evidence of Sa lower respiratory tract infection in the 14 days prior to enrollment. In the pilot cohort, 1 of 15 potential subjects was excluded for prior Sa infection, yielding 14 total subjects. In the validation cohort, 5 of 83 potential subjects were excluded for prior Sa infection and 2 potential subjects were excluded because AN specimens could not be collected within 24 hours of enrollment, yielding 76 total subjects. All enrolled pilot and validation subjects provided AN specimens at enrollment.

  • Clinical Data Collection: Using the Penn Data Store, a repository of clinical data compiled from HUP’s electronic medical record, we measured subject demographics, including age, sex, race, and ethnicity; underlying medical diagnoses; antibiotic exposure prior to enrollment; and baseline laboratory values. Clinical respiratory culture data, antibiotic administration data, and vital sign data were captured for 30-days from subject enrollment.

  • Specimen Collection, Data Collection, and Processing: AN specimens were collected by placing a sterile, flocked swab approximately 2cm inside the anterior nares and contacting the anterior mucosal surface for approximately 30 seconds. ET specimens were collected via the inline suction catheter of the subject’s endotracheal tube. The catheter was first flushed with approximately 5 mL of sterile saline, then advanced into the distal trachea and approximately 5 mL of sterile saline was flushed into the trachea and suctioned back into a Lukens trap. AN and ET specimens were stored immediately on ice and transferred to −80°C storage within 60 min of collection. After DNA extraction (QIAGEN DNeasy Powersoil) and amplification of the V1-V2 16S ribosomal RNA (rRNA) gene hypervariable region (27F - 338R) as previously described [17], we performed paired-end 250bp sequencing (Illumina MiSeq), sequence demultiplexing and alignment (QIIME2), sequence denoising and amplicon sequence-variant (ASV) binning (DADA2 with trimming to bases 13-200). ASVs were used to permit granular analysis and to improve compatibility across studies [18, 19]. Genus-level taxonomic assignments were performed by QIIME 2’s default classifier (Project SILVA_132_SSURef_Nr99 database) [20, 21]. We defined ASVs with high-confidence taxonomic assignment to Sa by BLAST alignment to the same SILVA database, with an e-value threshold of 1e-10 (Supplemental Figure 1).

  • Definition of Outcome: Lower respiratory Sa was defined as the proportional abundance of reads from ASVs with high-confidence Sa taxonomic assignment (linear model). Sa LRTI (binomial model) was defined as the detection of Sa by clinical tracheal aspirate culture (i.e., ordered by clinical care team given suspected infection), in addition to recorded fever (temperature > 100.4°F) or increased/new antibiotic orders within 48 hours of clinical culture order, within 30 days of enrollment.

  • Causal Models: We investigated two causal models to explain Sa LRTI: (1) based on abundance of intranasal Sa ASVs alone, and (2) based on non- Sa intranasal bacteria, conditional on the abundance of key Sa ASVs and prior antibiotic exposure.

  • Statistical Methods: Data were organized using R statistical software version 3.6.1 [22], and plots generated using the “ggplot2” package [23]. Potential confounders were compared between exposure groups using Wilcoxon rank-sum testing (continuous variables) and Fisher’s exact test (categorical variables). The impact of pre-enrollment antibiotics was assessed with joint species distribution models, using the “jSDM” package, but model convergence was poor and results are not reported [24, 25]. To permit partial pooling of parameter estimates across ASVs, regularization of ASV effects, and rigorous evaluation of model fit, Bayesian mixed-effects models were fit using Stan Hamiltonian Monte Carlo (HMC) version 2.19 [26–29]. Models were fit with 4 chains of 1000 iterations, confirmed with HMC diagnostics (no divergent iterations, no iterations saturating maximum tree depth, Rhat statistic < 1.1 for all parameters, and E-BFMI > 0.2), and by examining the posterior distributions [30]. Regularized horseshoe priors were used to encode the assumption of sparsity among ASV effects and control for false discovery [31]. Univariable models were fit using high-confidence Sa ASVs alone (random effects); multivariable mixed effects models were fit using non- Sa ASVs (random effects) and key Sa ASVs (fixed effects), as well as all recent antibiotic exposures (fixed effects). Binomial models also included an adjustment for variation in per-specimen sequencing depth. Figures highlight parameter estimates associated with ASVs with posterior probabilities of type S error (adjusted for false discovery via regularization, as above) < 0.25, < 0.1, and < 0.05 [32, 33]. Binomial models were initially fit using scaled ASV abundance data [34]. To further account for the possibility of confounding by compositionality, analyses were repeated after applying the centered log-ratio (CLR) transformation [35].

  • Power and Sample Size: Based on the number of Sa LRTI cases, as well as the precision of the observed risk associated with Sa ASVs and the protective effects associated with a Corynebacterium ASV in the pilot cohort, we estimated the necessary validation cohort size [36]. Approximately 70 subjects would yield odds ratio certainty intervals with the precision to permit detection of single ASV effects with type S error < 5%; we targeted enrollment of 10% more subjects to allow for a margin of error in that estimate.

  • Secondary Analyses: Linear models were fit to evaluate the relationship between AN and ET Sa ASVs, restricted to paired, same-day AN and ET specimens. All 14 pilot cohort subjects provided both AN and ET specimens. All validation cohort subjects provided AN specimens, but ET specimens were available for only 23 of them for logistical reasons. (No significant difference was observed in the AN microbiome between validation subjects with or without ET specimens; PERMANOVA/adonis analysis [37] of AN community variance attributable to whether ET specimens could be collected: unweighted Jaccard R2 0.013, unweighted UniFrac R2 0.008, weighted Jaccard R2 0.012, weighted UniFrac R2 0.039).

  • Availability of Data: Sequence data is publicly available on the National Center for Biotechnology Information’s Sequence Read Archive (NCBI SRA) with BioProject ID: PRJNA529220. Model code, as well as code used to produce the manuscript and figures, is available at github.com/bjkellyio.


Results:

  • Clinical characteristics are similar between critically ill subjects with and without detectable intranasal Sa: Baseline characteristics were similar between the exposed (detectable AN high-confidence) Sa ASVs at enrollment) and unexposed (no detectable AN high-confidence Sa ASVs at enrollment) groups in both cohorts, with no statistically significant differences identified in Sa-active antibiotic exposure in the 7 days prior to enrollment (Table 1). In the validation cohort, subjects with detectable intranasal Sa were slightly younger with median (IQR) age 59 (48, 67) years versus 64 (57, 71) years). 2 (4.9%) carried the diagnoses of leukemia or lymphoma in the Sa group, versus 13 (37%) in the group without detectable Sa. In the validation cohort, a slightly greater baseline serum creatinine and serum white blood cell (WBC) count were observed among subject with detectable intranasal Sa (Wilcoxon rank-sum p = 0.0048 and p = 0.045, respectively). We observed Sa LRTI events within 30 days of enrollment in 2 (14%) of pilot cohort subjects, and in 5 (6.6%) of validation cohort subjects.

  • Intranasal bacterial community diversity is similar across Sa exposure and outcome groups: We found no significant difference in bacterial community diversity (Shannon index, base e) between exposure groups in the pilot or validation cohorts: median (IQR) Shannon diversity 2.66 (2.29, 4.43) among pilot subjects with detectable Sa, and 3.32 (2.22, 3.69) among pilot subjects without (Wilcoxon p-value 0.749; 2.47 (2.08, 3.44) among validation subjects with detectable Sa, and 2.51 (1.79, 3.02) among validation subjects without (Wilcoxon p-value 0.48). We likewise found no significant difference in intranasal bacterial community diversity between subjects with and without Sa LRTI within 30 days of enrollment: Shannon diversity (median (IQR)) 1.75 (1.61, 1.88) in pilot subjects with Sa LRTI, 3.41 (2.56, 3.96) in pilot subjects without Sa LRTI (Wilcoxon p-value 0.068); Shannon diversity was 2.45 (2.36, 2.46) among validation subjects with Sa LRTI, and 2.51 (1.87, 3.26) among validation subjects without Sa LRTI (Wilcoxon p-value 0.88).

  • Sa ASVs have a bimodal distribution in the anterior nares and lower respiratory tract: Heatmaps of family-level taxonomic assignment demonstrated the persistent presence of AN Staphylococcaceae and Corynebacteriaceae, and suggested an association between upper and lower respiratory Staphylococcaceae across the total 228 AN and 94 ET specimens collected (Supplemental Figure 2A, 2B). At the ASV-level, we observed longitudinal persistence of subject-specific Staphylococcus and Corynebacterium sequence variants over time (Supplemental Figure 2C, 2D). The proportional abundance of Sa (total 16S rRNA gene reads assigned to high-confidence Sa ASVs divided by all 16S rRNA gene reads per specimen) was bimodally distributed across all AN and ET specimens (Figure 1A). To understand the relationship between Sa proportional abundance at the two sites, we examined all same-day pairs of anterior nares and endotracheal specimens (a total of 90 same-day paired AN-ET specimens from 14 pilot and 22 validation cohort subjects). We found a linear relationship between proportional abundance at the two sites, albeit with some outliers (low endotracheal proportional abundance despite high anterior nares proportional abundance; Figure 1B).

  • A subset of Sa ASVs account for high Sa proportional abundance in the anterior nares and lower respiratory tract: Among 8 Sa ASVs detected across the paired specimens, four were present at high abundance at both AN and ET sites (Figure 1C). We corroborated this observation by modeling the impact of anterior nares Sa ASV proportional abundance on total lower respiratory Sa proportional abundance. Models were fit independently in each cohort and across the pooled subjects (Supplemental Figure 3). Across all cohorts, we found 4 high-confidence Sa ASVs with intranasal proportional abundance positively correlated with total ET Sa proportional abundance (posterior type S probability < 0.05).

  • Intranasal Sa ASVs associated with high lower respiratory tract Sa abundance are also associated with increased odds of Sa LRTI: Given the observed association between specific intranasal Sa ASVs and increased total lower respiratory Sa, we evaluated the impact of intranasal Sa ASVs on the outcome of Sa LRTI within 30 days of enrollment and found three of the same Sa ASVs were associated with increased odds of Sa LRTI (Figure 2A). Three of the ASVs had high posterior certainty associations with Sa LRTI risk in the pilot study; only one association reached high posterior certainty in the validation cohort; but all three had very high posterior certainty associations, with type S error (posterior probability of null or negative effect) < 0.05, in the pooled data. Odds ratio (50% posterior credible interval) of 30-day Sa LRTI was 1.2 (1.1 - 1.2) for the first (ASV ID: f011d78941f0c011aca75252ef604d17), 3 (1.9 - 4.8) for the second (ASV ID: adcde8c78396022660fd6204471c65c8), and 3 (2 - 4.7) for the third (ASV ID: 241b2b5f1ea0f1065ca7027534068198). For comparison, we also evaluated the Sa LRTI risk associated with lower respiratory tract Sa ASVs detectable on the first day of enrollment (enrollment ET specimens available for all 14 pilot subjects and 15 validation cohort subjects) and found that the same three high-risk Sa ASVs were associated with increased Sa LRTI risk.

  • In the presence of high-risk intranasal Sa ASVs, intranasal Corynebacterium species reduce odds of Sa LRTI: Having established the risk of Sa LRTI associated with specific intranasal Sa ASVs, we evaluated the impact of the rest of the intranasal microbiome on Sa LRTI. To avoid simply identifying non- Sa ASVs associated with the three high-risk Sa ASVs that we had already identified, we modeled the relationship between intranasal non- Sa ASVs (random effects) and Sa LRTI while conditioning on the abundance of the three high-risk Sa ASVs, as well as prior antibiotic exposure (fixed effects). In the pilot cohort, we found a single protective ASV, assigned to Corynebacterium jeikeium. In the validation cohort, we again found a protective Corynebacterium jeikeium ASV, as well as protective ASVs assigned to three other Corynebacterium species (Figure 2B). In summary, we found that among 597 non- Sa intranasal ASVs, 16 had effects with high posterior certainty (i.e., type S error < 0.1), and 9 had effects with very high posterior certainty (i.e., type S error < 0.05) in either the pilot cohort, validation cohort, or pooled subjects. Among the high posterior certainty ASV effects, 5 ASVs assigned to the genera Staphylococcus, Neisseria, and Cutibacterium, were associated with increased odds of Sa LRTI. Conversely, 11 ASVs were associated with reduced odds of Sa LRTI, even after adjusting for high-risk Sa ASVs and prior antibiotic exposure. 2 of the protective ASVs were assigned to the genus Staphylococcus, 1 to the genus Pseudomonas, and 8 (72.7%) of these were assigned to the genus Corynebacterium. To account for potential confounding by compositionality [35], we repeated this analysis after CLR transformation of ASV abundance data. Of 8 protective Corynebacterium ASVs, all remained protective, and 6 with protective effects achieving high posterior certainty, after CLR transformation.

  • The impact of non- Sa ASVs on risk of Sa LRTI align with species-level ASV assignments: After excluding high-confidence Sa ASVs, 3 of the 69 other Staphylococcus ASVs had high posterior certainty associations with increased odds of Sa LRTI. Two of these three had high-confidence taxonomic assignments to Staphylococcus haemolyticus, while species level assignment for the third could not be resolved with high confidence. Of two non- Sa Staphylococcus ASVs strongly associated with decreased odds of Sa LRTI, one had high-confidence taxonomic assignment to Staphylococcus epidermidis and the other’s species level assignment could not be resolved with high confidence. In contrast, the impacts of intranasal Corynebacterium ASVs were less species-specific. Among the 8 ASVs assigned to the genus Corynebacterium that demonstrated high posterior certainty (including 4 with very high posterior certainty), all were associated with decreased odds of Sa LRTI. These 8 ASVs included two ASVs with high-confidence assignments to Corynebacterium jeikeium; two with high-confidence assignments to Corynebacterium accolens; one each with high-confidence assignments to Corynebacterium propinquum and Corynebacterium kroppenstedtii; and two ASVs for which species-level assignment could not be resolved with high confidence. Thus, multiple Corynebacterium species and sequence variants within the AN microbiome are linked to reduced risk of Sa LRTI even accounting for the presence of Sa (Supplemental Figure 4).


Discussion:

High-throughput sequencing has demonstrated the capacity for early detection of Sa LRTI [17, 38, 39]. We extended this method to understand the relationship between the nasal microbiota and lower respiratory tract microbiota during mechanical ventilation and how interspecific interactions impact Sa infection risk, focusing specifically on nasal microbiota and Sa LRTI because nasal microbiota can be assayed non-invasively and it is known that intranasal Sa colonization increases the risk of Sa infection [3, 6]. Recent analysis of the intranasal microbiome have suggested that interactions between Sa and Corynebacterium species are important to Sa colonization itself [11, 12], but it was unknown whether Corynebacterium or other genera impact Sa infection risk after accounting for its effect on Sa colonization. We report here that specific intranasal Corynebacterium species reduce the risk of Sa LRTI among critically ill patients with intranasal Sa colonization. Our results lay the foundation for novel risk stratification approaches and, potentially, therapeutic interventions, which might target the entire intranasal microbiome, rather than just Sa colonization.

We found a strong association between intranasal and lower respiratory Sa proportional abundance, and we confirmed an association between intranasal Sa abundance and risk for Sa LRTI. We demonstrated that these observed associations were driven by specific Sa 16S rRNA gene ASVs, which were linked to both greater lower respiratory Sa abundance and increased odds of Sa LRTI. Our analysis of non- Sa ASVs corroborated previous reports of interactions between Sa and Corynebacterium species, but importantly showed an impact not just on Sa colonization but also on Sa LRTI. Of 11 intranasal ASVs identified with high confidence as protective against Sa LRTI (even after adjusting for the abundance of high-risk Sa ASVs and prior antibiotic exposure), 8 (72.7%) were assigned to the genus Corynebacterium.

Corynebacterium species have previously been demonstrated to attenuate Sa virulence in vitro. In particular, Sa grown with Corynebacterium striatum demonstrates decreased transcription of virulence genes [15]. But the impact of Corynebacterium - Sa interactions on longitudinal risk for respiratory tract infections was previously unexplored [16]. Our study is the first to demonstrate the clinical impact of Corynebacterium interactions with Sa in the nares, demonstrating reduced odds for Sa ventilator-associated LRTI associated with increased abundance of specific Corynebacterium ASVs, even in the presence of high-risk Sa ASVs. We further demonstrated the impact of non- Sa Staphylococcus species. Staphylococcus epidermidis has previously been shown to protect against Sa colonization [40]. In our cohorts, a Staphylococcus epidermidis ASV was associated with decreased odds of Sa LRTI, even in the presence of high-risk Sa ASVs. Conversely, two Staphylococcus haemolyticus ASVs were associated with increased odds of Sa LRTI.

Our study has several limitations. Though we enrolled independent pilot and validation cohorts, both were from the same site, thus leaving uncertain the generalizability of our findings. We employed established sequence denoising methods to mitigate high-throughput sequence read errors, and applied mixed effects models with partial pooling of ASV effects to further account for potential misclassification, but our detection of nasal bacterial community members relied upon the 16S rRNA gene sequencing approach, which, unlike shotgun metagenomic analysis, may bias ascertainment of certain community members. Likewise, taxonomic assignments based on 16S rRNA gene sequencing only permit high-confidence assignment of certain reads and are contingent upon the chosen taxonomic database. Finally, we did not evaluate the bacterial community of the oral cavity, raising the question of whether oral Sa, Corynebacterium species, and other bacteria may mediate the observed association between nasal and lower respiratory bacterial communities.

In summary, our findings advance the understanding of how intranasal Sa colonization impacts risk for invasive Sa infection during critical care and bolster evidence for the importance of Sa - Corynebacterium interspecific interactions. We demonstrate for the first time the direct impact of key intranasal Corynebacterium ASVs on the odds of Sa LRTI. These associations should be further validated in new populations at risk for Sa LRTI, and explored in populations at risk for other invasive Sa infections. Work must be done to better resolve the identity of intranasal and respiratory tract Corynebacterium species, and to investigate the mechanisms of interactions between specific Corynebacterium species and Sa. Nevertheless the study’s findings lay the foundation for novel risk stratification tools, which incorporate more complete measures of the intranasal microbiome, and raise the possibility of novel therapeutic interventions, which might mitigate risk for Sa respiratory tract infection by modifying the entire intranasal microbiome, rather than just intranasal Sa.


Acknowledgements:

BJK is supported by the National Institute for Allergy and Infectious Diseases (K23 AI121485 and L30 AI120149), as well as Centers for Disease Control and Prevention (CDC) contract awards (BAA 200-2016-91964 and 200-2018-02919). BJK and EL are supported by the CDC Healthcare-Associated Infection Prevention Epicenters Program (U54CK000485). RGC and FDB are supported by R33HL137063. We acknowledge financial and other support from the Penn-CHOP Microbiome Program and the Penn Center for AIDS Research (P30AI045008).


Tables:

Table 1:

Table 1: Subject characteristics. Subject demographics, laboratory values at day of enrollment, medical comorbidities, and recent antibiotic exposures are presented for the pilot and validation cohorts. For categorical variables, counts and proportions are given; for continuous variables, medians and interquartile ranges are given. Subjects in both cohorts are compared by whether Sa was detected by AN 16S rRNA gene sequencing at enrollment. Categorical values are compared with Fisher’s exact test; continuous variables are compared with Wilcoxon rank-sum tests. (WBC: white blood cell count; AST: aspartate aminotransferase; ALT: alanine aminotransferase; COPD: chronic obstructive pulmonary disease; ILD: interstitial lung disease).


Figures:

Figure 1:

Figure 1: Relationship between Staphylococcus aureus proportional abundance in the anterior nares and lower respiratory tract. (A) Across all AN and ET specimens collected, we observed a bimodal distribution of Sa proportional abundance. (B) Relationship between ET Sa proportional abundance and AN Sa proportional abundance among a subset of all pairs of specimens that were collected on the same day (90 pairs across 36 subjects). Each dot represents an ET-AN specimen pair; the black line depicts the prediction of a linear model fit to the pairs, and the gray band depicts the model’s posterior certainty intervals. (C) The same comparison between ET and AN specimen pairs, but here depicting the proportional abundance of each detected Sa ASV, rather than aggregate Sa proportional abundance. As above, each dot represents a specimen pair, the line depicts a linear model fit to the pairs, and the gray band shows the model’s posterior certainty interval.


Figure 2:

Figure 2: Relationship between Staphylococcus aureus LRTI observed within 30-days of enrollment and anterior nares microbiome features on enrollment. (A) Parameter estimates from a mixed effects binomial model evaluating only the 9 high-confidence Sa AN ASVs observed across the pilot and validation subjects on the day of enrollment. The horizontal axis depicts the odds ratio of Sa LRTI on a logarithmic scale. The points indicate parameter medians, and the horizontal lines indicate parameter 50% posterior credible intervals. The broken vertical line above 1 indicates the null effect (odds ratio of 1). The color indicates parameters for which the posterior type S error is < 0.05, < 0.1, or < 0.25. (B) The parameter estimates from a mixed effects binomial model evaluating non- Sa ASVs (random effect), while adjusting for known high-risk Sa ASVs and prior antibiotic exposure (fixed effects). The figure only includes ASVs with a posterior type S error < 0.1; the color indicates parameters for which the posterior type S error (which accounts for multiple comparisons) is < 0.05, < 0.1, or < 0.25.


Supplemental Figures:

Supplemental Figure 1:

Supplemental Figure 1: High-confidence Staphylococcus aureus taxonomic assignment from 16S rRNA gene sequence data. The top panel depicts the count of all ASVs with genus-level assignment to Staphylococcus, and the species to which they were assigned. The bottom panel depicts the same, but with assignments shown as a proportion of all Staphylococcus ASVs.


Supplemental Figure 2:

Supplemental Figure 2: Bacterial community composition in the anterior nares and endotracheal aspirates of critically ill subjects dependent upon mechanical ventilation. Heatmaps depict the proportional abundance of (A) bacterial families across 266 anterior nares specimens and (B) bacterial families across 123 endotracheal aspirate specimens. The anterior nares bacterial community is further characterized by heatmaps depicting (C) all Staphylococcus ASVs and (D) all Corynebacterium ASVs. Specimens are arranged from left to right in order of collection; vertical white lines separate subjects.


Supplemental Figure 3:

Supplemental Figure 3: Relationship between total lower respiratory tract Staphylococcus aureus proportional abundance and the proportional abundance of Staphylococcus aureus sequence variants detected in the anterior nares. Parameter estimates from a mixed effects linear model are depicted for the 8 Sa ASVs observed across pilot and validation subjects with paired (same day) AN and ET specimens. The horizontal scale is fold change in ET Sa proportional abundance. The points indicate parameter medians, and the horizontal lines indicate parameter 50% posterior credible intervals. The broken vertical line above 0 indicates the null effect. The color indicates parameters for which the posterior type S error (which accounts for multiple comparisons) is < 0.05, < 0.1, or < 0.25.


Supplemental Figure 4:

Supplemental Figure 4: Phylogenetic relationships between Staphylococcus and Corynebacterium ASVs detected in the anterior nares. (A) Staphylococcus ASVs are depicted, with label color indicating the median posterior estimate of associated odds ratio of Sa LRTI (univariable model). (B) Corynebacterium ASVs are depicted, with label color indicating the median posterior estimate of associated odds ratio of Sa LRTI after adjustment for high-risk intranasal Sa ASVs, as well as prior antibiotic exposure (multivariable model).


References

1. Willyard C. The drug-resistant bacteria that pose the greatest health threats. Nature 2017; 543: 15.

2. Dangerfield B, Chung A, Webb B, Seville MT. Predictive value of methicillin-resistant staphylococcus aureus (MRSA) nasal swab PCR assay for MRSA pneumonia. Antimicrobial Agents and Chemotherapy 2014; 58: 859–864.

3. Lowy FD. Staphylococcus aureus infections. The New England Journal of Medicine 1998; 339: 520–532.

4. Solberg CO. A study of carriers of staphylococcus aureus with special regard to quantitative bacterial estimations. Acta Med Scand Suppl 1965; 436: 1–96.

5. Sollid JUE, Furberg AS, Hanssen AM, Johannessen M. Staphylococcus aureus: Determinants of human carriage. Infection, Genetics and Evolution 2014; 21: 531–541.

6. Wertheim HFL, Melles DC, Vos MC, Leeuwen W van, Belkum A van, Verbrugh HA, et al. The role of nasal carriage in staphylococcus aureus infections. The Lancet Infectious Diseases 2005; 5: 751–762.

7. Hurley JC. World-wide variation in incidence of staphylococcus aureus associated ventilator-associated pneumonia: A meta-regression. Microorganisms 2018; 6.

8. Hurley JC. Unusually high incidences of staphylococcus aureus infection within studies of ventilator associated pneumonia prevention using topical antibiotics: Benchmarking the evidence base. Microorganisms 2018; 6.

9. Kalanuria AA, Ziai W, Mirski M. Ventilator-associated pneumonia in the ICU. Critical Care 2014; 18: 208.

10. Stulik L, Hudcova J, Craven DE, Nagy G, Nagy E. Low efficacy of antibiotics against staphylococcus aureus airway colonization in ventilated patients. Clinical Infectious Diseases 2017; 64: 1081–1088.

11. Lemon KP, Klepac-Ceraj V, Schiffer HK, Brodie EL, Lynch SV, Kolter R. Comparative analyses of the bacterial microbiota of the human nostril and oropharynx. mBio 2010; 1.

12. Yan M, Pamp SJ, Fukuyama J, Hwang PH, Cho D-Y, Holmes S, et al. Nasal microenvironments and interspecific interactions influence nasal microbiota complexity and s. Aureus carriage. Cell Host & Microbe 2013; 14: 631–640.

13. Liu CM, Price LB, Hungate BA, Abraham AG, Larsen LA, Christensen K, et al. Staphylococcus aureus and the ecology of the nasal microbiome. Science Advances 2015; 1: e1400216.

14. Escapa IF, Chen T, Huang Y, Gajare P, Dewhirst FE, Lemon KP. New insights into human nostril microbiome from the expanded human oral microbiome database (eHOMD): A resource for the microbiome of the human aerodigestive tract. mSystems 2018; 3.

15. Ramsey MM, Freire MO, Gabrilska RA, Rumbaugh KP, Lemon KP. Staphylococcus aureus shifts toward commensalism in response to corynebacterium species. Frontiers in microbiology 2016; 7: 1230.

16. Bomar L, Brugger SD, Lemon KP. Bacterial microbiota of the nasal passages across the span of human life. Current Opinion in Microbiology 2018; 41: 8–14.

17. Kelly BJ, Imai I, Bittinger K, Laughlin A, Fuchs BD, Bushman FD, et al. Composition and dynamics of the respiratory tract microbiome in intubated patients. Microbiome 2016; 4: 7.

18. Bolyen E, Rideout JR, Dillon MR, Bokulich NA, Abnet C, Al-Ghalith GA, et al. QIIME 2: Reproducible, interactive, scalable, and extensible microbiome data science. 2018.

19. Callahan BJ, McMurdie PJ, Rosen MJ, Han AW, Johnson AJA, Holmes SP. DADA2: High-resolution sample inference from illumina amplicon data. Nature Methods 2016; 13: 581–583.

20. Yarza P, Richter M, Peplies J, Euzeby J, Amann R, Schleifer K-H, et al. The all-species living tree project: A 16S rRNA-based phylogenetic tree of all sequenced type strains. Systematic and Applied Microbiology 2008; 31: 241–250.

21. Munoz R, Yarza P, Ludwig W, Euzéby J, Amann R, Schleifer K-H, et al. Release LTPs104 of the all-species living tree. Systematic and Applied Microbiology 2011; 34: 169–170.

22. R Foundation for Statistical Computing R Core Team. R: A language and environment for statistical computing. 2019. Vienna, Austria.

23. Wickham H. Ggplot2: Elegant graphics for data analysis, 1st ed. 2009. Corr. 3rd printing 2010 edition. 2016. Springer International Publishing.

24. Warton DI, Blanchet FG, O’Hara RB, Ovaskainen O, Taskinen S, Walker SC, et al. So many variables: Joint modeling in community ecology. Trends in Ecology & Evolution 2015; 30: 766–779.

25. Brooks SP, Gelman A. General methods for monitoring convergence of iterative simulations. Journal of Computational and Graphical Statistics 1998; 7: 434–455.

26. McElreath R. Statistical rethinking: A bayesian course with examples in r and stan. 2015. Chapman; Hall/CRC.

27. Carpenter B, Gelman A, Hoffman MD, Lee D, Goodrich B, Betancourt M, et al. Stan: A probabilistic programming language. Journal of Statistical Software, Articles 2017; 76: 1–32.

28. Lee J, Sison-Mangus M. A bayesian semiparametric regression model for joint analysis of microbiome data. Frontiers in microbiology 2018; 9: 522.

29. Carney SM, Clemente JC, Cox MJ, Dickson RP, Huang YJ, Kitsios GD, et al. Methods in lung microbiome research. American Journal of Respiratory Cell and Molecular Biology 2019.

30. Gabry J, Simpson D, Vehtari A, Betancourt M, Gelman A. Visualization in bayesian workflow. J R Stat Soc A 2019; 182: 389–402.

31. Piironen J, Vehtari A. Sparsity information and regularization in the horseshoe and other shrinkage priors. Electronic journal of statistics 2017; 11: 5018–5051.

32. Gelman A, Tuerlinckx F. Type s error rates for classical and bayesian single and multiple comparison procedures. Computational statistics 2000; 15: 373–390.

33. Storey JD. A direct approach to false discovery rates. J R Stat Soc Series B Stat Methodol 2002; 64: 479–498.

34. Robinson MD, Oshlack A. A scaling normalization method for differential expression analysis of RNA-seq data. Genome Biology 2010; 11: R25.

35. Gloor GB, Macklaim JM, Pawlowsky-Glahn V, Egozcue JJ. Microbiome datasets are compositional: And this is not optional. Frontiers in microbiology 2017; 8: 2224.

36. Rothman KJ, Greenland S. Planning study size based on precision rather than power. Epidemiology 2018; 29: 599–603.

37. Anderson MJ. A new method for non-parametric multivariate analysis of variance: NON-PARAMETRIC MANOVA FOR ECOLOGY. Austral Ecol 2001; 26: 32–46.

38. Langelier C, Fung M, Caldera S, Deiss T, Lyden A, Prince BC, et al. Detection of pneumonia pathogens from plasma cell-free DNA. American Journal of Respiratory and Critical Care Medicine 2019.

39. Prazak J, Iten M, Cameron DR, Save J, Grandgirard D, Resch G, et al. Bacteriophages improve outcomes in experimental staphylococcus aureus ventilator-associated pneumonia. American Journal of Respiratory and Critical Care Medicine 2019; 200: 1126–1133.

40. Sullivan SB, Kamath S, McConville TH, Gray BT, Lowy FD, Gordon PG, et al. Staphylococcus epidermidis protection against staphylococcus aureus colonization in people living with human immunodeficiency virus in an inner-city outpatient population: A cross-sectional study. Open forum infectious diseases 2016; 3: ofw234.


R Session Information

## R version 3.6.3 (2020-02-29)
## Platform: x86_64-pc-linux-gnu (64-bit)
## Running under: Ubuntu 16.04.6 LTS
## 
## Matrix products: default
## BLAS:   /usr/lib/libblas/libblas.so.3.6.0
## LAPACK: /usr/lib/lapack/liblapack.so.3.6.0
## 
## locale:
##  [1] LC_CTYPE=en_US.UTF-8       LC_NUMERIC=C              
##  [3] LC_TIME=en_US.UTF-8        LC_COLLATE=en_US.UTF-8    
##  [5] LC_MONETARY=en_US.UTF-8    LC_MESSAGES=en_US.UTF-8   
##  [7] LC_PAPER=en_US.UTF-8       LC_NAME=C                 
##  [9] LC_ADDRESS=C               LC_TELEPHONE=C            
## [11] LC_MEASUREMENT=en_US.UTF-8 LC_IDENTIFICATION=C       
## 
## attached base packages:
## [1] stats     graphics  grDevices utils     datasets  methods   base     
## 
## other attached packages:
##  [1] patchwork_1.0.0  ggtree_1.16.6    vegan_2.5-6      lattice_0.20-41 
##  [5] permute_0.9-5    ape_5.3          phyloseq_1.28.0  cowplot_1.0.0   
##  [9] ggsci_2.9        ggnewscale_0.4.1 bayesplot_1.7.1  tidybayes_2.0.3 
## [13] gtsummary_1.3.0  gt_0.2.0.5       forcats_0.5.0    stringr_1.4.0   
## [17] dplyr_0.8.5      purrr_0.3.4      readr_1.3.1      tidyr_1.0.3     
## [21] tibble_3.0.1     ggplot2_3.3.0    tidyverse_1.3.0 
## 
## loaded via a namespace (and not attached):
##   [1] readxl_1.3.1        backports_1.1.7     plyr_1.8.6         
##   [4] igraph_1.2.5        lazyeval_0.2.2      splines_3.6.3      
##   [7] svUnit_1.0.3        crosstalk_1.1.0.1   rstantools_2.0.0   
##  [10] inline_0.3.15       digest_0.6.25       foreach_1.5.0      
##  [13] htmltools_0.4.0     rsconnect_0.8.16    fansi_0.4.1        
##  [16] magrittr_1.5        checkmate_2.0.0     cluster_2.1.0      
##  [19] Biostrings_2.52.0   modelr_0.1.7        matrixStats_0.56.0 
##  [22] xts_0.12-0          prettyunits_1.1.1   colorspace_1.4-1   
##  [25] rvest_0.3.5         haven_2.2.0         xfun_0.13          
##  [28] callr_3.4.3         crayon_1.3.4        jsonlite_1.6.1     
##  [31] lme4_1.1-23         survival_3.1-12     zoo_1.8-8          
##  [34] iterators_1.0.12    glue_1.4.1          gtable_0.3.0       
##  [37] zlibbioc_1.30.0     XVector_0.24.0      pkgbuild_1.0.8     
##  [40] Rhdf5lib_1.6.1      rstan_2.19.3        BiocGenerics_0.30.0
##  [43] scales_1.1.1        DBI_1.1.0           miniUI_0.1.1.1     
##  [46] Rcpp_1.0.4.6        viridisLite_0.3.0   xtable_1.8-4       
##  [49] tidytree_0.3.3      StanHeaders_2.19.2  stats4_3.6.3       
##  [52] DT_0.13             htmlwidgets_1.5.1   httr_1.4.1         
##  [55] threejs_0.3.3       arrayhelpers_1.1-0  ellipsis_0.3.1     
##  [58] farver_2.0.3        loo_2.2.0           pkgconfig_2.0.3    
##  [61] dbplyr_1.4.3        labeling_0.3        tidyselect_1.1.0   
##  [64] rlang_0.4.6         reshape2_1.4.4      later_1.0.0        
##  [67] munsell_0.5.0       cellranger_1.1.0    tools_3.6.3        
##  [70] cli_2.0.2           generics_0.0.2      ade4_1.7-15        
##  [73] broom_0.5.6         ggridges_0.5.2      evaluate_0.14      
##  [76] biomformat_1.12.0   fastmap_1.0.1       yaml_2.2.1         
##  [79] processx_3.4.2      knitr_1.28          fs_1.4.1           
##  [82] nlme_3.1-147        mime_0.9            rstanarm_2.19.3    
##  [85] xml2_1.3.2          compiler_3.6.3      shinythemes_1.1.2  
##  [88] rstudioapi_0.11     treeio_1.8.2        reprex_0.3.0       
##  [91] statmod_1.4.34      stringi_1.4.6       ps_1.3.3           
##  [94] Matrix_1.2-18       nloptr_1.2.2.1      markdown_1.1       
##  [97] shinyjs_1.1         multtest_2.40.0     vctrs_0.3.0        
## [100] pillar_1.4.4        lifecycle_0.2.0     BiocManager_1.30.10
## [103] data.table_1.12.8   httpuv_1.5.2        R6_2.4.1           
## [106] promises_1.1.0      gridExtra_2.3       IRanges_2.18.3     
## [109] codetools_0.2-16    boot_1.3-25         colourpicker_1.0   
## [112] MASS_7.3-51.6       gtools_3.8.2        assertthat_0.2.1   
## [115] rhdf5_2.28.0        withr_2.2.0         shinystan_2.5.0    
## [118] S4Vectors_0.22.1    mgcv_1.8-31         parallel_3.6.3     
## [121] hms_0.5.3           grid_3.6.3          minqa_1.2.4        
## [124] coda_0.19-3         rvcheck_0.1.8       rmarkdown_2.1      
## [127] Biobase_2.44.0      shiny_1.4.0.2       lubridate_1.7.8    
## [130] base64enc_0.1-3     dygraphs_1.1.1.6